For most of email marketing history, "deliverability" meant one thing: getting your message past the spam filter. If it reached the inbox, you won. That era is over. Today, AI decides whether your email goes to the inbox, lands in the Promotions tab, appears as a one-line summary, or gets ignored entirely — and it does this before any human has a chance to read it.
Today, three separate AI systems — at Gmail, Apple, and Microsoft — are making editorial decisions about your emails after they arrive. They're deciding whether your campaign appears in the Primary tab or gets buried in Promotions. Whether it shows up as a one-line summary the reader may never click. Whether it's pre-collapsed and hidden behind a "show more" link. Or whether it's quietly deprioritized into a category most people check once a week.
Understanding these systems has become as important as writing a compelling subject line. Here's a complete breakdown of what each one does — and how to work with them.
Gmail's Categorization AI: Primary vs. Promotions
Gmail introduced its tabbed inbox in 2013, but the AI behind it has evolved dramatically. What started as simple rules based on sender headers and list-unsubscribe signals is now a deep-learning system trained on billions of user interactions.
The core question Gmail asks about every incoming email: "Is this something the user personally wrote to, or is it a broadcast?" If it looks like a broadcast — even a useful, opted-in one — it routes to Promotions by default.
The signals that push emails into Promotions include:
- List-Unsubscribe header: This is the single biggest trigger. Any email that includes this header (which all major ESPs add automatically for compliance) is flagged as bulk mail. You can't remove it — and you shouldn't, since it's legally required — but it means Gmail treats your message as commercial from the start.
- Sending infrastructure: Emails sent via Mailchimp, Klaviyo, HubSpot, or any shared ESP domain are recognized immediately. Gmail cross-references the sending IP and domain against known marketing platforms.
- HTML structure: Heavy use of tables, banners, buttons, and tracking pixels signals promotional intent. Plain-text emails — or HTML that closely resembles plain text — are more likely to reach Primary.
- Image-to-text ratio: High ratios push toward Promotions. A single hero image with minimal text reads as an ad.
- Link patterns: Multiple CTAs, tracking-wrapped URLs, or UTM parameters are all strong signals that the email is designed for mass engagement rather than personal communication.
- Sender engagement history: This is where individual user behavior matters. If a specific subscriber has previously opened, replied to, or moved your emails to Primary, Gmail will start routing future emails there for them. Conversely, if they never open, you drift toward Promotions — or worse.
"Gmail's AI isn't trying to hurt your campaigns. It's trying to predict what the user wants to see first. The problem is that 'useful commercial email' and 'bulk promo' look nearly identical to an algorithm."
The Promotions Tab Isn't Dead — But It's Different
One important nuance: landing in Promotions doesn't mean you've failed. Gmail has invested in making the Promotions tab a shopping-friendly surface with rich previews, deal highlights, and expiry date annotations. For e-commerce sends with clear offers, Promotions can actually be a better placement than Primary because users arrive there in a browsing mindset.
The problem is for B2B and service businesses. A SaaS trial reminder or a consultancy's case study email doesn't benefit from being next to coupon codes. These emails need Primary placement — and that requires a fundamentally different content strategy.
Apple Intelligence: Priority, Summary, and the Collapsed View
Apple rolled out its AI-powered inbox management in iOS 18 and macOS Sequoia in late 2024, and by 2026 it's active on hundreds of millions of devices. Apple's system introduces three behaviors that email marketers need to understand:
1. Priority Messages
Apple Mail's AI surfaces a small selection of emails at the top of the inbox under a "Priority" section. These are predicted to be the most time-sensitive or personally relevant messages. The selection model weights heavily toward: direct replies in a conversation thread, emails from frequent senders, emails with personal names (not brand names) in the From field, and messages with explicit time references ("your meeting tomorrow," "your order ships today").
Marketing emails almost never surface as Priority. But transactional emails — order confirmations, account alerts, personalized reports — frequently do. This is a meaningful distinction for lifecycle email strategy.
2. AI Summaries
Apple's on-device AI generates single-line summaries for every email, which appear in the inbox list view in place of the preheader text you carefully crafted. The summary is written by the model based on the first 200-300 words of the email body — it ignores subject lines, preheaders, and anything in image alt text.
This has two consequences. First, your preheader no longer controls the inbox preview on Apple Mail — the AI does. Second, emails with complex HTML structures or heavy image use may generate generic or unhelpful summaries like "View this email in your browser" if the visible text content is thin.
The fix: Front-load meaningful, specific text in the email body. The first paragraph should read like a compelling news lead — not a decorative intro that only makes sense visually.
3. Notification Collapsing
Apple also groups and collapses repeated senders. If you send three emails in a week, Apple may show only the most recent one and hide the others behind "2 more emails from [Brand]." Subscribers may never realize the others exist.
Microsoft Outlook: Focused Inbox and Copilot Digests
Outlook's Focused Inbox applies a binary sort: Focused (important) vs. Other (everything else). The model is trained per-user and factors in: whether the sender is in the recipient's contact list, reply frequency, whether emails have ever been manually moved, and organizational signals (emails from colleagues rank higher than unknown senders).
More significantly, Microsoft 365 Copilot now offers an "Email Digest" feature for business users that summarizes multiple emails into a single AI-generated briefing. Your campaign may appear in someone's inbox as a bullet point in a digest — rewritten, condensed, stripped of branding. The CTA button doesn't survive the digest. The subject line doesn't either.
This means for B2B audiences using Microsoft 365 (which is a very large share of enterprise), your email's value proposition needs to survive being paraphrased into a single sentence.
The Signals Every AI Inbox Reads
Across all three platforms, a common set of signals emerges. These are the variables that most consistently influence inbox placement and AI treatment:
- Engagement velocity: How quickly and how often subscribers engage with your emails. A list that opens regularly builds positive sender-receiver history in every platform's model.
- From name vs. From address: "Alex from Acme" outperforms "Acme Newsletter" in nearly every test. Person names read as personal; brand names read as broadcast.
- Subject line signals: Phrases like "% OFF," "Limited time," "Deal," or excessive capitalization are flagged in all three systems. Natural-language subjects that could plausibly come from a colleague perform better.
- Text content density: Emails with substantial, readable text content score better than emails that are primarily images or buttons. The AI needs text to understand what the email is about — and to summarize it.
- Sending frequency consistency: Irregular spikes trigger review. Consistent cadences build algorithmic trust.
- List hygiene: Sending to a large proportion of unengaged subscribers drags down your sender reputation across all platforms. Inactive segments should be regularly suppressed.
How to Optimize for the AI-Curated Inbox
The practical implication of all this is that email marketing strategy is bifurcating. The same campaign approach can't serve a B2B SaaS audience on Outlook and an e-commerce audience on Gmail. You need to think in terms of audience-platform pairs.
Here's what works across all scenarios:
- Write the opening paragraph as if it's the only text the reader will see. Apple will summarize it. Copilot will paraphrase it. Gmail users may only glance at the preview. Make the first 50 words carry the full message.
- Personalize the From name. Sending from a person's name rather than a brand name is one of the highest-leverage moves available to marketers. It affects Primary placement, Priority scoring, and perceived relevance simultaneously.
- Segment aggressively and suppress early. The engaged portion of your list actively helps your sender reputation. The unengaged portion actively hurts it. A smaller, cleaner send beats a large, indiscriminate one every time.
- Use plain-text or near-plain-text for B2B audiences. Particularly for prospecting, reactivation, and high-stakes nurture sequences. The visual flair costs you Primary placement.
- Write for humans first, algorithms second. The AI systems — especially Apple's summarizer and Copilot — are trying to detect genuine value. Emails that are genuinely specific, personal, and relevant are harder to downgrade than templates. An email that feels like it was written specifically for this reader, about their actual situation, is the hardest thing for any AI to classify as bulk.
The Deeper Shift: AI Reading Before Humans Do
The bigger picture here is a structural change in how email works. For 30 years, email was a protocol — messages moved from server to inbox, and humans decided what to do with them. Now there are AI intermediaries between the send and the read, and their judgment shapes what gets seen.
This isn't going to reverse. Google, Apple, and Microsoft are all investing heavily in making their AI inboxes better at filtering signal from noise. The next generation of inbox AI will be more aggressive about collapsing, summarizing, and deprioritizing bulk sends — not less.
The marketers who will perform best in this environment are the ones who understand that email is no longer just a delivery problem. It's a relevance problem. The question isn't whether your message arrived. It's whether the algorithm decided it was worth a human's attention — before the human ever had a chance to decide for themselves.
"In the AI inbox era, deliverability is not a technical issue. It's a content and strategy issue. The algorithm is asking the same question your reader would ask: 'Does this email have something genuinely worth my time?'"
What This Means for AI-Generated Email Campaigns
One important implication for teams using AI tools to generate email campaigns: the quality of the output matters more than ever. An AI that generates generic, template-heavy, CTA-dense emails will produce content that gets downgraded by inbox AI. An AI that generates specific, narrative-driven, genuinely personalized email copy — built around real brand voice and real audience insight — produces content that looks to an inbox algorithm the same way it looks to a reader: like something worth reading.
This is the real competitive advantage of using AI not just to automate but to think: building campaigns where the strategy, the copy, and the design are all optimized not just for clicks, but for the AI gatekeepers that now stand between your send and your reader's attention.